A Study on Fatigue among Drivers with Machine Learning by Using CNN and SVM

A major contributing factor to traffic accidents is driver fatigue brought on by long stretches of driving, lack of sleep, boredom, or any combination of these. Therefore, it has been suggested that fatigue detecting systems be used to warn or alert drivers. But how early driver fatigue is detected frequently affects how well the system works. Conventional techniques aim to identify driver fatigue in real time; however, in numerous crucial situations, like the takeover transition phase in fully automated driving, this detection may come too late. Therefore, the objective of this work is to predict the driver's transition from nonfatigue to fatigue during highly automated driving using physiological indications. First, we used the ground truth for driver fatigue, which is PERCLOS, or the percentage of time the eyelids are closed. Subsequently, we selected the physiological features that were most important for anticipating driver weariness in advance. Finally, we used these crucial physiological characteristics to create prediction models that, by applying a method called nonlinear autoregressive exogenous network, could forecast the exhaustion transition at least 13.8 seconds in advance. The recommended method's potential was demonstrated by the accuracy of tiredness transition prediction for highly automated driving (F1 measure = 97.3 percent and a score of 99.1 percent for two types of models, respectively).